HAL Id: hal-02379240
https://hal.archives-ouvertes.fr/hal-02379240
Submitted on 12 Dec 2019HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci-entific research documents, whether they are pub-lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Environmental Structure: Evidence from a Visual
Associative Learning Task
Laurie-Anne Sapey-Triomphe, Sandrine Sonié, Marie-Anne Henaff, Jeremie
Mattout, Christina Schmitz
To cite this version:
Adults with autism tend to underestimate the hidden environmental
structure: evidence from a visual associative learning task
Laurie-Anne Sapey-Triomphe1, Sandrine Sonié1,2,3, Marie-Anne Hénaff1, Jérémie Mattout1*, Christina Schmitz1*
1. Lyon Neuroscience Research Center, Brain Dynamics and Cognition team, INSERM UMRS 1028, CNRS UMR 5292, Université Claude Bernard Lyon 1, Université de Lyon, F-69000, Lyon, France 2. Centre de Ressource Autisme Rhône-Alpes, Centre Hospitalier Le Vinatier, Bron, France. 3. Hôpital Saint-Jean-de-Dieu, Lyon, France.
* These authors contributed equally to this work
Abstract
The learning-style theory of Autism Spectrum Disorders (ASD) (Qian and Lipkin 2011) states that individuals with ASD differ from neurotypics in the way they learn and store information about the environment and its structure. ASD would rather adopt a lookup-table strategy (LUT: memorizing each experience), while neurotypics would favor an interpolation style (INT: extracting regularities to generalize). In a series of visual behavioral tasks, we tested this hypothesis in 20 neurotypical and 20 ASD adults. ASD participants had difficulties using the INT style when instructions were hidden but not when instructions were revealed. Rather than an inability to use rules, ASD would be characterized by a disinclination to generalize and infer such rules.
Keywords: autism; perception; categorization; learning; local and global processing
Corresponding author:
Correspondence concerning this article should be addressed to Dr. Christina Schmitz Email : [email protected]
INTRODUCTION
An atypical learning style was mentioned in the very first reports about autism (Asperger 1944; Kanner 1943). Léo Kanner had noticed that children with autism could not learn from adults in “conventional ways” (Kanner 1943). For instance, one child with autism had “an unusual memory for faces and names, knew the names of a great number of houses”, but “seemed unable to generalize, transfer an expression to another similar object” (Kanner 1943). Hans Asperger also described individuals with autism as being poor at “mechanical
learning” (Asperger 1944; Frith 1991), referring to the learning style spontaneously used by
typically developing children.
In Autism Spectrum Disorders (ASD), peculiarities of learning often suggested a failure to adapt rules and to generalize (Dawson et al. 2005; Plaisted 2001). Particularly, many studies have underlined an atypical category learning in ASD, usually slower and less accurate than in neurotypical (NT) participants (Alderson-Day and McGonigle-Chalmers 2011; Carmo et al. 2017; Church et al. 2010, 2015; Gastgeb et al. 2012; Gastgeb and Strauss 2012; Klinger and Dawson 2001; Soulières et al. 2007, 2011; Vladusich et al. 2010). The ability to extract regularities or common features between different stimuli in order to categorize them has often been investigated in ASD by using dot pattern categorization tasks. Studies using such tasks showed different degrees of impairment in children (Church et al. 2010) and adults (Gastgeb et al. 2012) with ASD, (although see Vladusich et al. 2010). Noticeably, individuals with ASD were more impaired than NT to categorize dot patterns that were most distorted from the category prototype (Froehlich et al. 2012). In ASD, studies showed an impaired ability to build up a mental representation of a category (Church et al. 2010; Klinger and Dawson 2001).
mechanisms could be at the heart of ASD symptoms, including the social ones. And the later could simply be most prominent since social stimuli are multidimensional, complex, essential and ubiquitous in our daily lives. This strongly motivates the study of such mechanisms, independently of social contexts.
Several theories have attempted to explain which mechanisms fail to be used in ASD, yielding atypical perception and learning. The Reduced generalization model (Plaisted 2001) suggested that individuals with ASD would have difficulties to extract similarities between stimuli, hence to generalize. Not only should this affect perception, but also social comprehension. Indeed, spontaneously processing the main underlying regularities of social stimuli or social situations will help generalizing (e.g. facial expressions) or using social norms in a context-dependent manner. More recently, a learning-style theory of autism was introduced (Qian and Lipkin 2011), inspired by the observation that individuals with ASD show difficulties to learn based on training examples (Dawson and Mottron 2008). The authors oppose two learning styles: the interpolation (INT) and the lookup-table (LUT) one. They suggest that the INT learning-style would be preferentially used by NT individuals, while individuals with ASD would be more biased toward the LUT learning-style.
are meaningless. In contrast, the INT learning style is context-dependent and particularly efficient in noisy environments, since encoding stimuli with broad tuning functions enables the categorization and interpretation of new stimuli. In other words, contrary to the LUT learning style, the INT one prevents from overfitting and would thus be more adapted to real-life situations where all kinds of noise have to be filtered out. This is essential to correctly interpret sensory stimuli, and subsequently to elaborate functions such as language (Fisher et al. 2014; Marcus et al. 2007) or appropriate social skills (Weston and Turiel 1980). Social stimuli are particularly noisy, flexible and context-dependent, and their multidimensional and complex underlying regularities need to be interpolated between situations. Hence, the INT style would be more optimal than the LUT style for processing social stimuli.
A wide variety of atypical behaviors in ASD could be explained by a reduced use of the INT style and a greater use of the LUT style (as compared to NT). For instance, some persons with ASD report that they tend to learn each situation almost by rote (LUT style) instead of generalizing (INT style). Temple Grandin explained: “When I encounter a new
social situation, I have to search my memory for a similar experience that I can use as a model for my next action. […] For common social interactions with clients I use preprogrammed, prerehearsed responses” (Grandin 1997). With the LUT style, accumulating
inconsistent regarding the degree of impairment (e.g. Soulières et al. 2007; Vladusich et al. 2010), and the ability to categorize in real-life ecological situations (where individuals with ASD are not given instructions) might differ from the situations of experimental testing described in scientific reports.
In the present study, we designed and tested a new visual paradigm questioning whether ASD individuals would indeed spontaneously make less use of the INT style and more use of the LUT styles than NT. Our paradigm matched the recommendation by Qian and Lipkin (2011) suggesting that age- and IQ-matched NT and ASD participants needed to be “trained on random (but fixed) association tasks and tasks with hidden, underlying rules”, and that it was “best to use non-social tasks (e.g., learning visual categorization of shapes) to
avoid potential confounds from autistic and typical subjects’ different developmental and intervention histories”.
Here, following Qian and Lipkin (2011), we hypothesized that contrary to NT participants, ASD participants would spontaneously favor a LUT over an INT learning strategy. Nevertheless, we also hypothesized that ASD participants would actually be able to understand and implement an INT strategy when explicitly instructed to do so, as the level of instructions can highly modify performance in ASD (Koldewyn et al. 2013; Van der Hallen et al. 2016). In other words, we hypothesized that in the case of limited guidance and minimal instructions, ASD participants would naturally favor a LUT learning strategy. However, we further hypothesized and tested that this difference with NT participants would diminish or even vanish in the case of explicit instructions about the strategy to be implemented.
perform (1) less accurately than NT in rule-based tasks, and (2) more accurately than NT in memory-based tasks. Upon completion of the series of learning tasks, participants completed questionnaires designed to elicit learning strategies used. We also expected that debriefing questionnaires would confirm that ASD participants were spontaneously biased toward the LUT style, whereas NT participants would be biased toward the INT style. Finally, participants were then asked to perform similar tasks but with clear instructions about the learning style to favor, in order to control for their ability to understand and implement each learning style. We here expected the behavior of ASD participants to resemble very much the one of NT participants and show that they were indeed able to implement one strategy or the other, as requested.
MATERIAL AND METHODS
Participants
Twenty-two participants with ASD and 23 NT participants took part in the study. Two participants with ASD and three NT participants were discarded from the analyses as they failed at performing the control task (see Results section). The two resulting groups consisted of 20 participants with ASD (mean age in years: 33.6 ±10.0) and 20 NT participants (mean age in years: 30.8 ±6.9). The two groups were matched for age, gender ratio and education level. Groups were also matched for intellectual quotient assessed by the Wechsler Adult Intelligence Scale (WAIS III or IV) on the verbal comprehension and perceptual reasoning subscores. Their characteristics are detailed in Table 1. Participants completed the autism-spectrum quotient (AQ) test (Baron-Cohen et al. 2001; Sonié et al. 2011). All participants scored above the cut-off threshold for ASD using the Autism Diagnostic Observation
only. Scores at the Autism Diagnostic Interview (Le Couteur et al. 2003) could be obtained for four participants only, due to the fact that many participants were diagnosed when they were adults. Based on these clinical assessments that concluded to a diagnosis of ASD also using criteria defined in the Diagnostic and Statistical Manual of Mental Disorders DSM-IV (American Psychiatric Association 2000) or DSM-V (American Psychiatric Association 2013), every participant then underwent an interview with an experienced psychiatrist highly specialized in autism diagnosis, research-certified in the ADI and the ADOS, and in charge of the regional Resources Center for Autism, before being enrolled in the study. All participants with ASD had a diagnosis of ASD without any intellectual deficiency or language acquisition delay. Participants had normal or corrected-to-normal vision. NT participants reported no history of neurological or psychiatric disorders. Approval was obtained from the local ethics committee (French South East IV Committee for the Protection of Persons). Participants gave their written consent beforehand.
--- Please, insert Table 1 here ---
Stimuli
We created visual stimuli that were unique and distinguishable enough to be memorized independently, but which could also be categorized based on common features. Importantly, these stimuli were new to all participants, so as to avoid any influence from prior knowledge. All stimuli were created using Matlab 2013a.
Each stimulus consisted of a geometric picture defined by lines and angles that formed a shape. It was made of nine points pseudo-randomly chosen on a horizontal ellipse and then connected with each other. The resulting pattern was filled with black color. Then, 104 of the ellipses were rotated by 15° [9:21], 32 by 90° [84:96], 104 by 135° [129:141], 32 by 255° [249-261] and 32 by 0° [-6:6] (Figure 1). These rotations enabled the formation of five
the orientation within each category with a standard deviation of +/- 6° in order to assess whether this would affect performance.
--- Please, insert Figure 1 here ---
General procedure
Participants were sitting at 60cm from the computer screen and the displayed stimuli were 20cm long and 8cm wide. Participants responded with their dominant hand, using the two buttons of the computer mouse. The experiments were programmed using the software package Presentation (Neurobehavioral Systems).
In each task, there were no more than five consecutive trials with the same winning response, and the same stimulus never appeared twice consecutively. Participants were asked to favor accuracy over speed. Participants completed six tasks and the whole experiment
lasted for about one hour. The order of the tasks was the following: 1st (or 2nd): Rule-based
task, 2nd (or 1st): Memory-based task, 3rd: Mixed LUT-INT task, 4th: Recognition test, 5th (or
6th): Rule-based control task, 6th (or 5th): Memory-based control task.
Main tasks without instructions Rule-based task
This task assessed the spontaneous use of the INT learning-style. Participants were only told that the displayed stimulus was informative about the side of the winning response. The hidden associative rule was such that the correct response depended on stimulus
orientation (Figure 1.B). Two categories of stimuli were presented: C0 (32 different horizontal
shapes) and C90 (32 different vertical shapes). The stimulus-winning-response association was
counter-balanced over participants. For half of the participants, a correct left button press (respectively, right) was associated with horizontal stimuli (respectively, vertical) and the reverse rule applied for the other half. As each stimulus was never presented twice, this association could not be memorized. The Rule-based task consisted of 64 trials with 64 different visual stimuli, and lasted for six minutes on average.
Memory-based task
This task assessed the spontaneous use of the LUT learning-style. Participants were only told that there was a link between the stimulus displayed and the side of the winning response. To succeed, participants had to memorize a unique association between a stimulus
and a response side (Figure 1.C). Eight stimuli from category C255 were pseudo-randomly
the four others were associated with a right winning answer. Associations were counter-balanced over participants. The Memory-based task consisted of 64 trials with 8 different visual stimuli, and lasted for six minutes on average.
We counterbalanced the order of presentation of the Rule-based task and
Memory-based task between participants.
Mixed LUT-INT task
This task tested the spontaneous use of both the LUT and INT learning strategies. Again, participants were told that there was a link between the stimulus displayed and the side of the winning response. They were also told that if they had discovered strategies during the two previous tasks (Rule-based and Memory-based tasks), these could be useful for this new task. To succeed, participants had to use the INT strategy for half of the stimuli (categorization), and the LUT strategy for the other half (memorization) (Figure 1.D). This task was divided into four sessions consisting of 64 trials each. In each session, 32 stimuli had
to be classified according to their orientations (16 belonging to C15, and 16 belonging to C135)
and 8 stimuli belonging to C255 were displayed four times each (i.e. 32 trials) and had to be
memorized. The four sessions presented a total of 64 stimuli from C15 (each presented once),
64 stimuli from C135 (each presented once), and 8 stimuli from C255 (each presented 16 times).
The Mixed LUT-INT task consisted of 256 trials in total, with 136 different visual stimuli. It lasted for about 24 minutes (interrupted by two-minute breaks in-between sessions).
Control tasks with full instructions Rule-based control task
stimulus examples illustrating the two different orientations. The design of this control task was the same as in the main Rule-based task. Two categories of stimuli were presented: 32
stimuli belonging to C15 (e.g., associated with a right winning response), and 32 belonging to
C135 (e.g., associated with a left winning response). The side of the winning response was
counter-balanced between participants. The Rule-based control task consisted of 64 trials with 64 different visual stimuli, and lasted for six minutes on average.
Memory-based control task
This task measured the participants’ performance when explicitly told to memorize the stimulus-winning-response association. The design of this control task was the same as in the
main Memory-based task. Eight stimuli belonging to C255 were pseudo-randomly presented
eight times each (these were different from the ones presented in the previous tasks). Half of the stimuli were associated with a right winning answer, and the other half with the left winning answer. The side of the winning answer was counter-balanced between participants. The Memory-based control task consisted of 64 trials with 8 different visual stimuli, and lasted for six minutes on average.
Additional measures Recognition test
This task tested whether participants did memorize some of the stimuli presented during the Mixed LUT-INT task, performed right beforehand. Note that this is different from having memorized the association between a given stimulus and the side of the winning response for that stimulus. Participants were presented with stimuli displayed during the
Mixed LUT-INT task or never displayed before, and they had to indicate by a button press
Sixteen stimuli from each of the three following categories were used: C15, C135 and
C255. Within each category, eight stimuli had been displayed during the Mixed LUT-INT task,
while the eight others were new. Each stimulus appeared for three seconds, and participants had to press on the side of the screen indicating “seen” or “never seen” (this side was counterbalanced between participants). No feedback was provided, and the inter-stimulus trial lasted for 800ms. The Recognition test consisted of 48 trials with 48 different visual stimuli (24 had been previously shown), and lasted for about 4 minutes.
Post-experiment questionnaires
After the fourth task (before starting the control tasks), participants completed a questionnaire. They were asked to report the strategies they tried to use for each task. We classified answers to the question “Can you describe the strategy you used for each of the exercises?” within four categories: categorization based on the orientation, on another rule based on global stimulus features (e.g.: large vs. small colored surface), on a rule based on local stimulus features (e.g.: side of the smallest angle of the shape), or on memorization.
Statistical analyses
Chance levels were calculated as the upper limit of the confidence interval in binomial tests assessing 50% of success (with a confidence level of 0.95), i.e. 68.1% of correct answers for 32 trials, 62.8% for 64 trials, 59.0% for 128 trials and 56.3% for 256 trials. Proportions of participants scoring above chance level or reporting different strategies in the questionnaires were compared using proportion tests.
Statistical analyses were performed using R (http://www.R-project.org). The threshold
for statistical significance was always set to p < .05.
RESULTS
No effect of orientation blurring within each stimulus categories could be observed on accuracy. Hence all the results reported below integrate over this dimension.
Participant selection criteria in the control tasks
--- Please, insert Table 2 here ---
Main tasks without instructions Rule-based task
Results are presented in Table 2 and Figure 2-A. The NT group performed better than the ASD group (p < .005), with 81.9% (±18.1) versus 62.8% (±16.6) of correct answers. A higher proportion of NT than ASD participants scored above chance level (NT: 75% vs. ASD: 25%, p < .05). As another indicator of group difference in inferring and applying the rule, 85% of the NT participants answered correctly eight times in a row after only 16 trials (median). In contrast, only 30% of the ASD participants reached the same performance, but after 27 trials (median).
--- Please, insert Figure 2 here ---
Memory-based task
Results are presented in Table 2 and Figure 2-B. The NT and ASD groups obtained 59.0% (±11.8) and 56.6% (±14.0) of correct answers, respectively. No differences in accuracy were found between groups. Both groups scored at chance level on average. Participants scoring above chance level represented 40% of the NT group and 30% of the ASD group (no group difference).
Mixed LUT-INT task
Results are presented in Table 2 and Figure 3. We performed a nested ANOVA investigating the effect of group (NT and ASD), stimulus (INT and LUT) and session (1 to 4) on accuracy. There was a group effect (F(38,1) = 7, p < .01), with the NT group scoring higher than ASD (NT: 71.3% ±11.2 vs. ASD: 62.3% ±13.4, p < .05). There was a stimulus
effect (F(38,1) = 43, p < 10-6), with a better recognition of INT stimuli than LUT stimuli
17, p < 10-6), with a significant increase from session 1 to session 3 (p < .01) and 4 (p < .01).
Finally, there was an interaction between the factors group and stimuli (F(38,1) = 20, p < 10
-4), with the NT group scoring higher than ASD for INT stimuli (NT: 81.5% ±16.4 vs. ASD:
64.3% ±19.9, p < .01), but not for LUT stimuli. Within group, NT participants scored higher for INT than LUT stimuli (p < .001), whereas ASD participants showed no difference in accuracy between INT and LUT stimuli.
Participants scoring above chance level for INT stimuli were more numerous in NT than ASD (NT: 90% vs. ASD: 40%, p < .001) and were not different for LUT stimuli (NT: 55% vs. ASD: 50%). To determine which learning style was associated with the highest accuracy, we compared accuracy for INT versus LUT stimuli for each participant (Figure 3B). The proportion of participants with higher scores for INT than LUT stimuli was larger in NT than ASD (NT: 85% vs. ASD: 30%, p < .001). The proportion of participants with higher scores for LUT than INT stimuli was larger in ASD than NT (NT: 15%, ASD: 50%, p < .05). In the ASD group, 20% of the participants did not show any differences between scores for INT and LUT stimuli.
--- Please, insert Figure 3 here ---
Control tasks with full instructions Rule-based control task
ASD participants committed twice more errors than NT participants between the 1st and 6th
trial.
Memory-based control task
Results are presented in Table 2 and Figure 2-B. The percentages of correct answers were 66.6% (±9.8) in NT and 61.9% (±14.5) in ASD, and did not significantly differ between groups. Participants scoring above chance level were 60% in the NT group and 55% in the ASD group (no group difference).
Between task comparisons
Rule-based vs. Memory-based tasks
In the main tasks without instructions, the NT group showed greater accuracy in the
Rule-based task than in the Memory-based task (p < 10-5). In the ASD group, there was no significant difference between these two tasks.
In the control tasks with instructions, both groups were less accurate in the
Memory-based control task than in the Rule-Memory-based control task (p < 10-6 in both groups).
Without vs. with full instructions
In the Rule-based tasks, giving instructions resulted in a significant increase in
accuracy by 16.5% (±18.4) in NT (p < .001) and 32.3% (±15.5) in ASD (p < 10-6) (Table 2).
In the Memory-based tasks, giving instructions resulted in a significant increase in accuracy by 7.6% (±12.3) in NT (p < .05) and in a non-significant increase by 5.2% (±16.1) in ASD.
Relationships with IQ scores
The working memory score (including attention and concentration assessments) was expected to correlate with accuracy in the Memory-based task. It proved to be the case for NT participants only (r = .49, p < .05).
Additional measures Recognition test
The overall accuracy in NT (60.9% ±8.2) and ASD (56.9% ±9.5) was not significantly different between groups. Groups did not differ in accuracy for LUT stimuli that needed to be memorized (NT: 70.9% ±13.9, ASD: 69.1% ±14.3), nor for INT stimuli that needed to be categorized (NT: 55.9% ±11.9, ASD: 50.8% ±14.4) (Figure 4A). Both groups remembered more accurately LUT stimuli than INT stimuli (p < .001 in both groups).
Post-experiment questionnaires
A higher proportion of NT than ASD participants reported searching for a rule in the
Rule-based task (NT: 100% vs. ASD: 60%, p < .01), in the Memory-based task (NT: 95% vs.
ASD: 50%, p < .01) and in the Mixed LUT-INT task (NT: 95% vs. ASD: 60%, p < .05) (Figure 4-B). In the Memory-based task, a higher proportion of ASD than NT participants tried to memorize the stimuli (ASD: 35% vs. NT: 5%, p < .05) (Figure 4-C).
Remarkably, the two ASD participants with the best accuracy for INT stimuli in the
Mixed LUT-INT task reported an alternative rule to the one based on orientation. They
classified the stimuli based on a single detail: given that the stimuli were angled, the highest
point of the shape was on the right side of the screen for stimuli belonging to C15 and on the
left side of the screen for stimuli belonging on C135. These two participants declared clicking
on the side of this highest point to succeed. They were the only two participants who reported having found this rule.
DISCUSSION
The goal of the present study was to investigate the spontaneous tendency of NT and ASD adults to use two kinds of learning styles: extracting regularities to interpolate between items (INT style) or memorizing lists of associations independently and precisely (LUT style). We had hypothesized that the ASD group would show a bias toward a decreased use of the INT style and an increased use of the LUT style, as compared to NT (Qian and Lipkin 2011). Two main results were in favor of this hypothesis: ASD participants were less inclined to spontaneously use the INT style and reported an increased tendency to use the LUT style. Importantly though, the ASD group was able to instantiate the INT style when instructed to do so.
Decreased use of the INT style and increased use of the LUT style in ASD
Qian and Lipkin (2011) had hypothesized that individuals with ASD would perform better with the LUT style than NT. Yet, in the present study, the ASD group did not score higher than NT in tasks requiring the use of the LUT style. Group differences might have emerged if a longer period of learning had been proposed. Indeed, although qualitative reports from questionnaires revealed that a higher proportion of participants with ASD than NT tried to use the LUT style, the Memory-based task was apparently too difficult for most participants to elicit a group difference. Further use of a similar task should probably consider a longer exposure to the stimuli in order to be more sensitive to a putative group difference in memorizing individual stimulus-outcome associations.
Noteworthy, there were positive correlations between accuracy with the INT style and the perceptual reasoning score (IQ-PR), and between accuracy with the LUT style and the working memory score (IQ-WM) within the NT group, but not within the ASD group. This suggests that, in the NT group at least, the participants’ abilities to find the orientation rule and to perform 3D visual search were related, and so were their abilities to memorize the association between stimuli and answers and their working memory scores. In the ASD group, their IQ scores could not be related to the intra-group variability in making use of the LUT or INT style.
Spontaneous versus instructed INT style
and local processing abilities in children with ASD in a free-choice task versus an instructed task (Koldewyn et al. 2013). Their study revealed that children with ASD did not show a disability in global processing when they received explicit instructions, but showed disinclination in global processing in absence of explicit instructions. Likewise, a recent study involving visual search tasks showed that children with ASD did not differ from typically developing children when given explicit instructions, but had a lower accuracy when they were less aware of the targets to search for (Van der Hallen et al. 2016). Inconsistent results about local and global processing (Van der Hallen et al. 2015, for a meta-analysis) could be explained by different levels of task instructions. Similarly, explicit instructions can also significantly reduce the impairment of ASD participants found in non-instructed tasks assessing social cognition or executive functions (Baez et al. 2012; Baez and Ibanez 2014; Senju et al. 2009; White et al. 2009). In their studies, group differences were explained by difficulties to spontaneously integrate social and/or contextual information in ASD (Baez et al. 2012; Senju et al. 2009). Altogether, results from the literature and from the present study highlight the key role of instructions in assessing abilities in ASD, which could account for inconclusive results in ASD (e.g. on categorization). It also suggests that highly structured and explicit rules or instructions are necessary for individuals with ASD.
Precision tuning in the learning-style theory
people with ASD had similar performance as NT (Bott et al. 2006). Other visual tasks have investigated perceptual abilities in ASD in contexts where one or several features needed to be encoded. For instance, people with ASD showed enhanced perceptual abilities to detect orientation in a simple grid, but impaired abilities to detect orientation when noise was added to the grid (Bertone et al. 2005). This suggests that, in ASD, visual processing at a low-level can be enhanced, but that it might impaired at a higher-level. In our tasks, a lower-level could correspond to tasks with instructions (where only the orientation needs to be encoded), and a higher-level could correspond to the non-instructed tasks (where several features, including noise, need to be encoded).
The learning-style theory of ASD is consistent with an older theory: the Adaptive Resonance Theory (Grossberg 1999). This theory stated that categorizing can be described as a top-down effect, decreasing differences within a category and increasing differences between categories. This top-down effect could adapt the precision tuning to allow categorization (i.e. with a broad precision tuning). In ASD, a decreased influence of top-down categorical knowledge on discrimination (Soulières et al. 2007) could explain the reduced use of the INT learning-style by ASD participants in our tasks.
From the learning-style theory to computational accounts of ASD
Computational approaches have recently enabled to shed light on how people with ASD cope with different types of uncertainty, namely sensory ambiguity, probabilistic
uncertainty and environmental uncertainty (Lawson et al. 2017; Palmer et al. 2017). These
three dimensions of uncertainty can be identified in our tasks. First, sensory uncertainty was manipulated as stimulus orientations were following a normal distribution. Yet, we did not find any impact of such a sensory noise on the difference in performance between groups. Second, probabilistic uncertainty was null, since each stimulus was surely predicting the correct response. What remained to be determined by the participant was the actual predicting rule at play. Finally, environmental uncertainty was introduced between blocks by changing the learning strategy, and within the Mixed LUT-INT task by mixing the type of stimulus-response association. We did observe a difference between groups in dealing with this environmental uncertainty, in the sense that only NT participants reported raising their strategy to the use of two learning-styles in the case of the Mixed LUT-INT task. However, this may highlight another difference between ASD and NT participants than the one reported in Lawson et al. (2017). Indeed, while Lawson and colleagues studied associative learning, we investigated both the learning of a simple (deterministic) association and the ability to discover the predicting feature in the cueing stimulus. This aspect and the finding that ASD participants proved impaired compared to NT will have to be further investigated. This should be studied from a computational perspective, in order to possibly identify common mechanisms that would also explain recent findings by Lawson et al. (2017), as well as findings about perceptual learning in probabilistic environments (Robic et al. 2014).
Limitations
tasks compared to the Rule-based ones, which yielded chance level performance on average in each group. This may have impaired our sensitivity to reveal group differences in memorization performance, expected to be in favor of ASD participants. The Memory-based tasks might have been too short for some participants to be able to learn the stimulus-outcome association.
Conclusion
We showed that people with ASD were less inclined to search for global underlying regularities, despite intact abilities to categorize. A reduced automatic and implicit learning of these regularities might impact daily life, and particularly, the understanding of social interactions (e.g., recognizing facial expressions requires being able to interpolate between facial expressions previously encountered). Studying the learning-styles in ASD can have concrete applications on learning strategies to use with children with ASD, and could explain, for instance, their difficulties to catch grammatical regularities and extract rules in general. Further studies investigating the spontaneous use of LUT and INT strategies in children with ASD may shed light on how these specificities unfold during the development.
Compliance with Ethical Standards
Funding: This study was supported by a Scientific Research Council grant from the
Vinatier Hospital Center, and was performed within the framework of the LABEX CORTEX (ANR-11-LABX-0042) of Université de Lyon, within the program "Investissements d'Avenir" (ANR-11-IDEX-0007) operated by the French National Research Agency (ANR).
Ethical approval: All procedures performed in this study were in accordance with the
References
Alderson-Day, B., & McGonigle-Chalmers, M. (2011). Is it a bird? Is it a plane? category use in problem-solving in children with autism spectrum disorders. Journal of autism and developmental disorders, 41(5), 555–565. doi:10.1007/s10803-010-1077-9
American Psychiatric Association. (2000). Diagnostic and Statistical Manual of Mental Disorders, 4th Edition. https://dsm.psychiatryonline.org/doi/abs/10.1176/appi.books.9780890420249.dsm-iv-tr. Accessed 18 December 2017
Asperger, Hans. (1944). Die ‘Autistischen Psychopathen’ im Kindesalter. Archiv für Psychiatrie und Nervenkrankheim, 117, 76–136.
Baez, S., & Ibanez, A. (2014). The effects of context processing on social cognition impairments in adults with Asperger’s syndrome. Frontiers in Neuroscience, 8, 270.
doi:10.3389/fnins.2014.00270
Baez, S., Rattazzi, A., Gonzalez-Gadea, M. L., Torralva, T., Vigliecca, N. S., Decety, J., et al. (2012). Integrating intention and context: assessing social cognition in adults with Asperger
syndrome. Frontiers in Human Neuroscience, 6, 302. doi:10.3389/fnhum.2012.00302
Baron-Cohen, S., Wheelwright, S., Robinson, J., & Woodbury-Smith, M. (2005). The Adult Asperger Assessment (AAA): a diagnostic method. Journal of Autism and Developmental Disorders, 35(6), 807–819. doi:10.1007/s10803-005-0026-5
Baron-Cohen, S., Wheelwright, S., Skinner, R., Martin, J., & Clubley, E. (2001). The autism-spectrum quotient (AQ): evidence from Asperger syndrome/high-functioning autism, males and
females, scientists and mathematicians. Journal of Autism and Developmental Disorders, 31(1), 5–17.
Bott, L., Brock, J., Brockdorff, N., Boucher, J., & Lamberts, K. (2006). Perceptual similarity in autism. Quarterly journal of experimental psychology (2006), 59(7), 1237–1254. doi:10.1080/02724980543000196
Carmo, J. C., Souza, C., Gonçalves, F., Pinho, S., Filipe, C. N., & Lachmann, T. (2017). Effects of categorical representation on visuospatial working memory in autism spectrum disorder. Journal of Clinical and Experimental Neuropsychology, 39(2), 131–141.
doi:10.1080/13803395.2016.1207754
Church, B. A., Krauss, M. S., Lopata, C., Toomey, J. A., Thomeer, M. L., Coutinho, M. V., et al. (2010). Atypical Categorization in Children with High Functioning Autism Spectrum Disorder. Psychonomic bulletin & review, 17(6), 862–868. doi:10.3758/PBR.17.6.862 Church, B. A., Rice, C. L., Dovgopoly, A., Lopata, C. J., Thomeer, M. L., Nelson, A., & Mercado, E.
(2015). Learning, plasticity, and atypical generalization in children with autism. Psychonomic Bulletin & Review, 22(5), 1342–1348. doi:10.3758/s13423-014-0797-9
Dawson, M., & Mottron, L. (2008). Learning in autism. In Learning and Memory: A Comprehensive Reference III Cognitive Psychology (J. Byrne and H. L. Roediger (Oxford: Elsevier)., pp. 759– 772).
Dawson, Michelle, Mottron, Laurent, & Gernsbacher, Morton Ann. (2005). Learning in Autism, Vol. [2] of Learning and Memory: A Comprehensive Reference, [759-772].
DSM-V, American Psychiatric Association. (2013). Diagnostic and Statistical Manual of Mental Disorders V. American Psychiatric Association.
Fisher, A. V., Godwin, K. E., Matlen, B. J., & Unger, L. (2014). Development of Category-Based Induction and Semantic Knowledge. Child Development. doi:10.1111/cdev.12277
Friston, K. (2003). Learning and inference in the brain. Neural networks: the official journal of the International Neural Network Society, 16(9), 1325–1352. doi:10.1016/j.neunet.2003.06.005 Frith, U. (1991). Autism and Asperger Syndrome. Cambridge University Press.
Gastgeb, H. Z., Dundas, E. M., Minshew, N. J., & Strauss, M. S. (2012). Category formation in autism: can individuals with autism form categories and prototypes of dot patterns? Journal of autism and developmental disorders, 42(8), 1694–1704. doi:10.1007/s10803-011-1411-x Gastgeb, H. Z., & Strauss, M. S. (2012). Categorization in ASD: The Role of Typicality and
Development. Perspectives on language learning and education, 19(2), 66–74. doi:10.1044/lle19.2.66
Grandin, T. (1997). A personal perspective on autism. In Handbook of Autism and Pervasive Developmental Disorders (Cohen, D.J. and Volkmar, F.R., eds) (pp. 1032–1042).
Grossberg, S. (1999). The link between brain learning, attention, and consciousness. Consciousness and Cognition, 8(1), 1–44. doi:10.1006/ccog.1998.0372
Kanner, Leo. (1943). Autistic disturbances of affective contact., Nervous Child, 2, 3, 217–250. Klinger, L. G., & Dawson, G. (2001). Prototype formation in autism. Development and
psychopathology, 13(1), 111–124.
Koldewyn, K., Jiang, Y. V., Weigelt, S., & Kanwisher, N. (2013). Global/local processing in autism: not a disability, but a disinclination. Journal of Autism and Developmental Disorders, 43(10), 2329–2340. doi:10.1007/s10803-013-1777-z
Lawson, R. P., Mathys, C., & Rees, G. (2017). Adults with autism overestimate the volatility of the sensory environment. Nature Neuroscience. doi:10.1038/nn.4615
Le Couteur, A., Lord, C., & Rutter, M. (2003). The autism diagnostic interview-revised (ADI-R) (Los Angeles: Western Psychological Services.).
Lord, C., Risi, S., Lambrecht, L., Cook, E. H., Leventhal, B. L., DiLavore, P. C., et al. (2000). The autism diagnostic observation schedule-generic: a standard measure of social and
communication deficits associated with the spectrum of autism. Journal of Autism and Developmental Disorders, 30(3), 205–223.
Ozonoff, S., & Miller, J. N. (1995). Teaching theory of mind: A new approach to social skills training for individuals with autism. Journal of Autism and Developmental Disorders, 25(4), 415–433. doi:10.1007/BF02179376
Palmer, C. J., Lawson, R. P., & Hohwy, J. (2017). Bayesian Approaches to Autism: Towards Volatility, Action, and Behavior. Psychological Bulletin. doi:10.1037/bul0000097
Plaisted, K. C. (2001). Reduced generalization in autism: An alternative to weak central coherence. In J. A. Burack, T. Charman, N. Yirmiya, & P. R. Zelazo (Eds.), The development of autism: Perspectives from theory and research (pp. 149–169). Mahwah, NJ, US: Lawrence Erlbaum Associates Publishers.
Qian, N., & Lipkin, R. M. (2011). A learning-style theory for understanding autistic behaviors. Frontiers in Human Neuroscience, 5, 77. doi:10.3389/fnhum.2011.00077
Robic, S., Sonié, S., Fonlupt, P., Henaff, M.-A., Touil, N., Coricelli, G., et al. (2014). Decision-Making in a Changing World: A Study in Autism Spectrum Disorders. Journal of Autism and Developmental Disorders. doi:10.1007/s10803-014-2311-7
Senju, A., Southgate, V., White, S., & Frith, U. (2009). Mindblind eyes: an absence of spontaneous theory of mind in Asperger syndrome. Science (New York, N.Y.), 325(5942), 883–885. doi:10.1126/science.1176170
Sonié, S., Kassai, B., Pirat, E., Masson, S., Bain, P., Robinson, J., et al. (2011). [French version of screening questionnaire for high-functioning autism or Asperger syndrome in adolescent: Autism Spectrum Quotient, Empathy Quotient and Systemizing Quotient. Protocol and questionnaire translation]. Presse Médicale (Paris, France: 1983), 40(4 Pt 1), e181-188. doi:10.1016/j.lpm.2010.07.016
Soulières, I., Mottron, L., Giguère, G., & Larochelle, S. (2011). Category induction in autism: slower, perhaps different, but certainly possible. Quarterly journal of experimental psychology (2006), 64(2), 311–327. doi:10.1080/17470218.2010.492994
Suzanne Scherf, K., Behrmann, M., Minshew, N., & Luna, B. (2008). Atypical development of face and greeble recognition in autism. Journal of Child Psychology and Psychiatry, 49(8), 838– 847. doi:10.1111/j.1469-7610.2008.01903.x
Van der Hallen, R., Evers, K., Boets, B., Steyaert, J., Noens, I., & Wagemans, J. (2016). Visual Search in ASD: Instructed Versus Spontaneous Local and Global Processing. Journal of Autism and Developmental Disorders, 46(9), 3023–3036. doi:10.1007/s10803-016-2826-1
Van der Hallen, R., Evers, K., Brewaeys, K., Van den Noortgate, W., & Wagemans, J. (2015). Global processing takes time: A meta-analysis on local-global visual processing in ASD.
Psychological Bulletin, 141(3), 549–573. doi:10.1037/bul0000004
Vladusich, T., Olu-Lafe, O., Kim, D.-S., Tager-Flusberg, H., & Grossberg, S. (2010). Prototypical category learning in high-functioning autism. Autism research: official journal of the International Society for Autism Research, 3(5), 226–236. doi:10.1002/aur.148 Weston, D. R., & Turiel, E. (1980). Act–rule relations: Children’s concepts of social rules.
Developmental Psychology, 16(5), 417–424. doi:10.1037/0012-1649.16.5.417
White, S. J., Burgess, P. W., & Hill, E. L. (2009). Impairments on “open-ended” executive function tests in autism. Autism Research: Official Journal of the International Society for Autism Research, 2(3), 138–147. doi:10.1002/aur.78
Tables
Table 1: Demographical and neuropsychological characteristics of the participants
NT group ASD group p
Number 20 20 ns
Male / Female 15 / 5 16 / 4 ns
Age (years) 30.8 (±6.9) 33.6 (±10.0) ns
Educational level (years) 4.6 (±2.6) 3.4 (±2.8) ns
WAIS IV Verbal comprehension 124 (±13) 123 (±18) ns Perceptual reasoning 111 (±14) 113 (±17) ns Working memory 112 (±11) 107 (±20) ns Processing speed 111 (±16) 105 (±25) ns AQ score 12 (±6) 35 (±8) < 10-6
Table 2: Group results of the behavioral tasks
Task NT (n=20) ASD (n=20) p Without instructions Rule-based accuracy 81.9% (±18.1) 62.8% (±16.6) ** % participants > chance 75% 25% * Memory-based accuracy 59.0% (±11.8) 56.6% (±14.0) ns % participants > chance 40% 30% ns
Mixed LUT-INT accuracy (Total) 71.3% (±11.2) 62.3% (±13.4) *
(INT) 81.5% (±16.4) 64.3% (±19.9) ***
(LUT) 61.1% (±15.8) 60.4% (±15.8) ns
% participants > chance (Total) 100% 60% **
(INT) 90% 40% *** (LUT) 55% 50% ns With instructions Rule-based accuracy 98.4% (±1.9) 95.2% (±6.3) * % participants > chance 100% 100% ns Memory-based accuracy 66.6% (±9.8) 61.9% (±14.5) ns % participants > chance 60% 55% ns Difference Rule-based accuracy 16.5% (±18.4) *** 32.3% (±15.5) *** ** Memory-based accuracy 7.6% (±12.3) * 5.2% (±16.1) * ns
Mean accuracy (± standard deviation) and percentage of participants scoring above chance level in each task in the NT and ASD groups. The “Difference” section provides the difference in accuracy between the tasks with instructions and the ones without. Reported p-values correspond to tests pertaining to accuracies or proportions of participants (* p < .05, **
p < .01, *** p < .001). The last column refers to between group comparisons.
Figures
Figure 1: Typical trial (left) common to most of tasks (right)
A. Example of trial presentation. This structure was the same in every task, except for the
Recognition test. ITI: intertrial interval.
B. The Rule-based task tested the spontaneous use of the INT strategy. Stimuli were
oriented along 0° (category C0) or 90° (category C90) and each stimulus only appeared
once. Stimuli had to be categorized according to their orientation.
C. The Memory-based task tested the spontaneous use of the LUT strategy. Stimuli were all
oriented along 255° (category C255) and were repeated eight times each. The stimulus /
winning response association had to be memorized.
D. In the Mixed LUT-INT task, C15 and C135 stimuli had to be classified according to their
orientation and were never repeated, whereas C255 stimuli had to be memorized and
appeared 16 times each.
Stimulus displayed 3 sec Won Options displayed up to 3 sec Option selected 800 ms Feedback 2000 ms ITI 500 ms B. Rule-based task INT C15 C. Memory-based task LUT C135 C255
D. Mixed LUT-INT task Categorization: orientation
C0 C90 C255 Memorization
Figure 2: Accuracy in the Rule-based and Memory-based tasks
A. Group mean accuracy observed in the Rule-based task without (left) and with instructions (right, control task).
B. Group mean accuracy observed in the Memory-based task without (left) and with instructions (right, control task).
NT group: blue, ASD group: orange. Error bars correspond to standard deviations. The dash
line indicates the chance level. # p < .05, * p < .01, ** p < .001, *** p < .0001.
*** # 0 25 50 75 100 Without
instructions instructionsWith
A c c ura c y (% ) 0 25 50 75 100 Without instructions With instructions A c c ura c y (% ) NT ASD
A. Rule-based tasks (INT style) B. Memory-based tasks (LUT style)
**
*
Figure 3: Accuracy in the Mixed LUT-INT task
A. Mean accuracy in the Mixed LUT-INT task for INT stimuli to be categorized (left) and LUT stimuli to be memorized (right).
** * A. 0 25 50 75 100
INT stimuli LUT stimuli
A c c ura c y (% ) NT ASD B. 0 25 50 75 100 NT ASD P e rc e nt a ge of pa rtic ipa nt s
> for INT stimuli No difference > for LUT stimuli
NT ASD 0 25 50 75 100 1 2 3 4 Session number NT ASD C. INT stimuli to be categorized
A c c ura c y (% ) 0 25 50 75 100 1 2 3 4 P e rc e nt a ge of pa rtic ipa nt s a nsw e ri ng a bov e c hanc e le v e l Session number NT ASD 0 25 50 75 100 1 2 3 4 Session number NT ASD A c c ura c y (% )
D. LUT stimuli to be memorized
B. Percentages of participants with a higher accuracy for INT stimuli than LUT stimuli (clear grey), with a higher accuracy for LUT stimuli than INT stimuli (black), or with no difference in accuracy between INT and LUT stimuli (grey). Numbers correspond to the number of participants for each category.
C. Results for INT stimuli across sessions: mean accuracy (left) and percentage of participants answering above chance level (right).
D. Results for LUT stimuli across sessions: mean accuracy (left) and percentage of participants answering above chance level (right).
Figure 4: Recognition test and questionnaire results
A. Percentage of correct answer in the Recognition test, for INT stimuli (C15 and C135)
and LUT stimuli (C255).
B-C. Percentage of participants who reported having used a rule-based strategy (B) or a
memory-based strategy (C) to perform the Rule-based task, Memory-based task, and
Mixed LUT-INT task (these tasks were performed without instructions). Participants
could report both strategies for a unique task. Error bars correspond to standard deviations. * p < .01
0 25 50 75 100 Rule-only
task Memo-onlytask LUT-INTtask NT ASD 0 25 50 75 100 Rule-only task Memo-only task LUT-INT task 0 25 50 75 100 P e rc e nt a ge of c orre c t re c ognit ion * * INT
stimuli stimuli LUT
A. B. C. Rule task Memo task LUT-INT
task Rule task Memo task
